Evaluating a person as kind or unkind, intelligent or unintelligent, etc., involves using

J Pers Soc Psychol. Author manuscript; available in PMC 2014 Dec 2.

Published in final edited form as:

PMCID: PMC4251455

NIHMSID: NIHMS578483

Abstract

Communicators, motivated by strategic self-presentation, selectively underreport negative content in describing their impressions of individuals and stereotypes of groups, particularly for targets whom they view ambivalently with respect to warmth and competence. Communicators avoid overtly inaccurate descriptions, preferring to omit negative information and emphasize positive information about mixed individual targets [Study 1]. With more public audiences, communicators increasingly prefer negativity omission to complete accuracy [Study 2], a process driven by self-presentation concerns [Study 3], and moderated by bidimensional ambivalence. Similarly, in an extension of the Princeton Trilogy studies, reported stereotypes of ethnic and national outgroups systematically omitted negative dimensions over 75 years—as anti-prejudice norms intensified—while neutral and positive stereotype dimensions remained constant [Study 4]. Multiple assessment methods confirm this stereotyping-by-omission phenomenon [Study 5]. Implications of negativity omission for innuendo and stereotype stagnation are discussed.

Keywords: negativity omission, stereotypes, racial and ethnic attitudes, warmth and competence, self-presentation

You’ve got to accentuate the positive,

Eliminate the negative,

Latch on to the affirmative,

Don’t mess with Mister In-Between!

As Bing Crosby exhorts in the classic song Ac-Cent-Tchu-Ate the Positive [Mercer, 1985], dwelling on negativity contravenes societal norms. Injunctions against derogating others date back at least to Biblical times: “As a north wind brings rain, so a sly tongue brings angry looks” [Prov. 25:23 New International Version], and the maxim “If you don’t have anything nice to say, don’t say anything at all” remains commonplace today. Our work probes communicators’ tendency, driven by self-presentation concerns, to strategically omit negativity when holding ambivalent [i.e., mixed-valence] information or stereotypes about other individuals or groups.

Conflicting Norms

Our research addresses the tension between anti-negativity and accuracy norms that arises when people seek to describe a target they view at least somewhat negatively. Accuracy maxims for cooperative conversation prescribe communication standards based on quality and quantity: truthfulness and completeness backed by sufficient evidence [Grice, 1975]. Audiences expect communicators to convey information honestly and without malice [Ekman, 2001], conveying “the truth, the whole truth, and nothing but the truth,” per the courtroom oath. Lying is usually condemned [Backbier, Hoogstraten, & Terwogt-Kouwenhoven, 1997] as morally and socially wrong [Bok, 1978]. Among 555 person descriptors, sincere and honest were rated most desirable, with liar and phony least likable [Anderson, 1968]. When accountable to audiences further along a communication chain, communicators seek to convey accurate information, to avoid blame for transmitting falsehoods [Lyons & Kashima, 2003].

Despite the apparent efficiency and rationality of candid communication, anti-negativity pressures are pervasive: Prescriptive norms call for politeness, protecting individuals’ public dignity or “face” [Brown & Levinson, 1987; Goffman, 1959]. Communicators more often use positive than negative words across all languages sampled [Boucher & Osgood, 1969]; this “Pollyanna principle” [Matlin & Stang, 1978] includes a person-positivity effect in evaluations of individuals, such that people are evaluated more favorably than inanimate targets [Sears, 1983]. Positivity pervades evaluations of real, fictive, known, and anonymous others [Feather & Armstrong, 1967; Gerard, 1961; Greenberg & Miller, 1966; Price, Harburg, & Newcomb, 1966].

Inappropriately criticizing individuals or groups may evoke negative responses not only from targets but also from audiences [Jones, Hester, Farina, & Davis, 1959; Sutton, Elder, Douglas, 2006]. Audiences dislike communicators who say they dislike others [Ames, Bianchi, & Magee, 2010; Folkes & Sears, 1977; Wyer, Budesheim, & Lambert, 1990]. This Transfer of Attitudes Recursively [TAR; Gawronski & Walther, 2008] arises when audiences infer that communicators enjoy criticizing others [Ames et al., 2010]. Spontaneous trait transference leads audiences to automatically associate communicators with qualities that they describe in others, particularly when criticized targets are absent [Skowronski, Carlston, Mae, & Crawford, 1998].

Inappropriately derogating groups, as opposed to individuals, poses still greater potential costs to communicators, eliciting more disliking—as well as attributions of prejudice—from audiences [Mae & Carlston, 2005]. Even audiences who agree with the biased remark respond negatively to communicators who derogate outgroup members [Mae & Carlston, 2005; Simon & Greenberg, 1996]. Current societal norms strongly condemn ethnic/national prejudice [Meertens & Pettigrew, 1997; Rogers & Prentice-Dunn, 1981], deeming racists as dislikable as terrorists, drunk drivers, and child molesters [Crandall, Eshleman, & O’Brien, 2002]. In sum, “nice people can’t be racists and racists can’t be nice people” [McConahay, 1986, p. 123].

Self-Presentation Concerns

Given the costs of violating societal anti-negativity norms by openly criticizing individuals or groups, omitting negativity in describing others makes strategic sense for communicators. As social beings, people need acceptance from others [Baumeister & Leary, 2005], leading them to select their words strategically when describing people [see Schaller & Conway, 1999]. Our reasoning parallels prior theorizing: “that people are motivated to create an attractive self-presentation is obvious […] in many instances a person would value a friendly relationship with another person more than communicating his true opinion. This may result in his biasing his opinions in a positive direction” [Folkes & Sears, 1977, p. 517] and “because the contents of speech influence impressions of speakers, individuals may be motivated by impression-management goals to engage in some selective self-censorship—to talk a lot about certain traits possessed by others and less so about other traits” [Schaller, Conway, & Tanchuk, 2002, p. 863]. Strategically presenting information about others constitutes an indirect self-presentation tactic, allowing communicators to manage impressions of the self [see Cialdini & Richardson, 1980].

For group targets, self-presentation typically involves concern with appearing prejudiced. Modern samples are highly motivated to control prejudice [Dunton & Fazio, 1997], for both external and internal reasons [Plant & Devine, 1998]. Self-reported personal prejudices correlate strongly with normative acceptability of prejudice against specific groups [r = .96; Bergsieker & Monin, 2006; Crandall et al., 2002], and manipulating perceived acceptability of prejudice alters self-reported prejudice [Blanchard, Crandall, Brigham, & Vaughan, 1994; Monteith, Deneen, & Tooman, 1996; Stangor, Sechrist, & Jost, 2001], suggesting that strategic self-presentation influences reported evaluations of groups.

Notably, we do not claim that all characterizations of others exclusively reflect strategic self-serving communication. Omission of negativity could arise from altruistic concerns about effects on the target, uncertainty about the target’s true nature, or—for outgroup stereotypes—genuine prejudice reduction. The extent of true prejudice reduction is difficult to assess [see Dovidio & Gaertner, 1986], however, because people learn to suppress inappropriate forms of prejudice [Crandall et al., 2002; Jackman & Muha, 1984]. Negativity omission may be multiply determined, but we argue that self-presentation [or impression management] concerns can drive communicators to omit negative content in describing impressions of individuals or groups.

Ambivalence

As negativity omission conflicts with accuracy norms, it may occur only under specific conditions. We assert that omission is most probable when communicators feel ambivalence toward the target. Rather than define ambivalence as conflicting evaluations of an object on a single evaluative dimension [see Kaplan, 1972], we focus on bidimensional ambivalence, a type of belief heterogeneity, defined as holding simultaneous oppositely valenced judgments of an object on two dimensions [Armitage, 2003]. Impressions of people—whether individuals or groups—typically contain two or more orthogonal dimensions, whose evaluative valence can be congruent or incongruent, reflecting either a halo effect [e.g., Thorndike, 1920] or bidimensional ambivalence [e.g., Glick & Fiske, 1996; Russell & Fiske, 2008]. In outgroup stereotypes, bidimensional ambivalence appears to be not the exception but the rule [Fiske, Cuddy, & Glick, 2007] and was evident a half-century ago: Karlins, Coffman, and Walters [1969] noted that each outgroup stereotype they studied comprised “both positive and negative terms. The consensus is never purely favorable or unfavorable, although the degree of evaluative differentiation within the stereotype varies widely” [p. 11]. Bidimensional ambivalence can lead to instability and amplification: People show variable, more extreme responses to ambivalently stereotyped targets [Gergen & Jones, 1963]. Unlike attitudes based on chiefly positive or negative evaluations, those displaying bidimensional ambivalence are less well formed or persistent [Armitage, 2003].

Bidimensional ambivalence toward targets may lead communicators to decide among possible responses based on the level of self-presentation concerns triggered by the situation. We theorize that ambivalence-related instability causes communicators’ expressed impressions of ambivalent targets to vary in valence—accentuating the positive, eliminating the negative—when self-presentation concerns mount, as when confronting a public audience or strong anti-prejudice norms. In contrast, positive or negative content in descriptions of univalent [uniformly positive or negative] targets should not vary substantially due to self-presentation concerns.

Negativity Omission and Stereotyping by Omission

We test two hypotheses regarding the tendency to omit negative information when describing individuals and groups. In the interpersonal domain, the negativity-omission hypothesis asserts that when describing individuals characterized by both positive and negative attributes, communicators will selectively report primarily positive information, preferring to omit negative information. Negativity omission for ambivalent targets allows communicators to comply with both “be accurate” and “don’t be mean” norms, avoiding costs of criticizing others. We theorize that self-presentation concerns drive omission, such that preferences for negativity omission over complete accuracy increase in contexts that raise self-presentation concerns.

Similarly, the stereotyping-by-omission hypothesis claims that communicators selectively omit negative and emphasize positive dimensions of ambivalent outgroup stereotypes. The terms negativity omission and stereotyping by omission distinguish between omission of negative information encoded about individuals, gained through direct experience, and negative aspects of group stereotypes, reflecting shared beliefs about groups that arise from socio-structural factors. Parallel to negativity-omission predictions, we suggest that self-presentation concerns, namely, concern with appearing prejudiced, motivate stereotyping by omission.

Our stereotyping-by-omission hypothesis provides an account of how and why reported ethnic and national group stereotypes have become more favorable over time [e.g., Meertens & Pettigrew, 1997], by positing that reports of historically unfavorable stereotypes systematically become neutral on negative dimensions, as anti-prejudice norms gain strength. According to this hypothesis, negative aspects of stereotypes do not readily reverse over time [e.g., shifting from “ignorant” to “intelligent”], but are instead omitted in favor of positive stereotypes from other domains [e.g., “passionate”]. Thus, increases in reported stereotype favorability may arise not only from genuine changes in stereotype content, but also strategic expression of only positive content for ambivalently stereotyped outgroups.

In sum, we assert that in describing ambivalent targets, anti-negativity norms supersede candor: Communicators eliminate negative content and accentuate the positive. They still value honesty—avoiding [subjectively] false information—but their descriptions fail to convey fully the positive and negative content in their impression of targets. Thus, communicators expressing their impressions of ambivalent individual targets or societal outgroup stereotypes prefer to convey the [subjective] truth—but only the positive parts of the truth—and nothing but the truth.

Warmth and Competence

Although bidimensional ambivalence and negativity omission can occur on numerous dimensions, we focus on warmth and competence, which constitute the two most fundamental dimensions of social cognition, predict affect and behavior, and often operate hydraulically. Success in navigating interpersonal interactions requires accurately inferring others’ warmth [i.e., morality] and competence, because they allow perceivers to encode respectively others’ helpful/harmful intent and ability/inability to enact said aims [Fiske et al., 2007]. In other words, warmth reflects the “interpersonal content of goals of an acting person—whether the goals are beneficial to other people and maintain moral norms, or are harmful to others and transgress the norms,” whereas competence reflects “efficiency in goal attainment—whether the goals are effectively reached or remain unattained” [Wojciszke, 2005, p. 165]. Warmth and competence are essential for self-preservation, are universal [Fiske et al., 2007], and explain over 80% of variance in perceptions of social behaviors [Wojciszke, Bazinska, & Jaworski, 1998]. Individuals and groups vary substantially in perceived warmth and competence [Fiske, Cuddy, Glick, & Xu, 2002; Russell & Fiske, 2008]. Bidimensional ambivalence comprises high warmth and low competence [e.g., traditional stereotypes of women; Glick & Fiske, 1996] or high competence and low warmth [e.g., stereotypes of Asians; Lin, Kwan, & Cheung, & Fiske, 2005], influencing whether perceivers pity, envy, help, or harm targets [Cuddy, Fiske, & Glick, 2007].

Warmth and competence can display a compensatory relationship. In comparative judgments of individuals and groups, perceivers infer that warm targets are relatively incompetent and that competent targets are relatively cold [Judd, James-Hawkins, Yzerbyt, & Kashima, 2005; Kervyn, Yzerbyt, & Judd, 2010]. In theory perceivers “might be perfectly content to maintain a negative view of a group on both competence and warmth if there exists a third dimension in which compensation is possible” [Judd et al., 2005, p. 910], but compensation effects have not yet emerged for other dimensions [e.g., healthiness; Yzerbyt, Kervyn, & Judd, 2008], suggesting that warmth and competence show a special compensatory relationship. We thus expect descriptions of ambivalent targets primarily to trade off warmth and competence—eliminating the negative dimension and accentuating the positive dimension—though we do not claim that omission is limited to warmth and competence. Communicators omitting negativity may also cite positive content on other dimensions [e.g., musicality, neatness, piety] when anti-negativity norms are particularly strong. However, to satisfy the Gricean quality and quantity maxims, descriptions should accurately convey targets’ warmth and competence, so an absent dimension most likely indicates omission.

The Present Research

Five studies test whether communicators omit negative content and emphasize positive content, as a function of self-presentation concerns, when they have ambivalent information or attitudes about the competence and warmth of target individuals or groups. The first studies test our individual-level negativity-omission hypothesis, clarifying the types of target individuals [Study 1], audiences [Study 2], and motives [Study 3] that lead to negativity omission. To test our group-level stereotyping-by-omission hypothesis, Study 4 examines reported stereotypes of 10 ethnic and national groups across 75 years of survey data collected using the Katz and Braly [1933] Princeton Trilogy adjective-checklist method. Study 5 replicates findings for current outgroup stereotypes using modern assessment methods. In sum, these studies probe a process by which communicators, mindful of self-presentation, selectively underreport negative content and emphasize positive content in describing ambivalent impressions of individuals and groups.

Study 1: Negativity Omission in Descriptions of Individuals

To investigate negativity omission with individual targets, we assessed communicators’ preferences for positive and negative descriptions of a target. We manipulated bidimensional ambivalence via the target’s warmth- and competence-related behavior, presented either as uniformly positive or negative [i.e., univalent] or as mixed—positive on one dimension and negative on the other [i.e., ambivalent].

Predictions

Given ambivalent information about a target, participants are hypothesized to prefer descriptions that selectively emphasize positive content and omit negative content, rather than describing the target completely accurately [or inaccurately]. Omission should be the dominant response only in the ambivalent conditions. For univalent positive targets, omission should not trump complete accuracy because both statement types are positively valenced and the latter also satisfies the completeness [quantity] maxim. For univalent negative targets, omission should not predominate because participants cannot substitute positive information without violating the quality maxim; instead, they should be no more inclined toward omission than complete accuracy or inaccuracy.

Method

Participants

We recruited 134 college students for course credit or a chance to win $10. Excluding 14 who failed a manipulation check, the final sample [N = 120] comprised 66 women and 54 men, with a mean age of 20.2, including 76 White and 44 non-White individuals.

Procedure and materials

In an “Impression Formation and Communication” online study, participants were randomly assigned to read one of four vignettes about a target person, whose behaviors varied systematically in terms of warmth and competence, then think about how they would describe this person. Next, participants rated their likelihood of making statements varying in warmth and competence to a casual acquaintance. A manipulation check tested memory for target behavior, followed by demographic questions.

The vignette described the target’s behavior as uniformly positive [intelligent and kind, n = 30], uniformly negative [unintelligent and unkind, n = 28], or ambivalent [intelligent but unkind, n = 32; unintelligent but kind, n = 30]. It read:

• Imagine someone named Pat, a student of your same age, class year, and gender who lived in your dorm and has taken several classes with you. In the course of getting to know each other, you have observed Pat making many [un]intelligent comments and [but] often treating other students [un]kindly.

Participants were told to read the vignette twice carefully, as they could not return to it later.

Next, participants saw eight statements characterizing the target person positively or negatively on warmth and competence in the form “Pat’s ___.” Four matched the vignettes: “smart and nice,” “smart but mean,” “nice but stupid,” or “mean and stupid.” The other four omitted one dimension, characterizing the target as “smart,” “nice,” “stupid,” or “mean.”1 Participants rated their likelihood of making each statement to a casual acquaintance on a scale from 1 [extremely unlikely] to 6 [extremely likely].

Statement classification

We classified the eight statements according to whether their characterization of the target was complete [and accurate], reflected omission on the expected dimension, or was inaccurate. For example, if the vignette described Pat’s behavior as kind but unintelligent, “nice but stupid” is completely accurate, “nice” indicates omission on the predicted dimension [i.e., negative content], and statements contradicting the vignette [e.g., “smart,” “nice and smart”] are inaccurate. For ambivalent targets, only statements omitting negative content constituted omission; supplemental analyses assessed statements omitting positive content. For univalent targets, we had no a priori basis for expecting systematic omission of warmth versus competence, so omitting either dimension counted as omission, including “smart” and “nice” for the intelligent/kind target and “stupid” and “mean” for the unintelligent/unkind target. We computed mean ratings for complete, omission, and inaccurate statements for each participant.

Notably, the predicted negativity-omission effect—communicators selectively omitting negative content and emphasizing positive content when describing ambivalent targets—lies in the contrast between completely accurate and omission statements. For ambivalent targets, negativity omission should manifest as rejecting complete statements, which include negative content, and selecting omission statements, which omit negative content and stress positive content. Thus, we operationalize negativity omission as preferring “smart” to “smart but mean” for intelligent/unkind targets and “nice” to “nice but stupid” for unintelligent/kind targets.

Results

Submitting statement ratings to mixed-factorial analysis of variance [ANOVA] revealed a significant 4 [target: intelligent/kind, intelligent/unkind, unintelligent/kind, unintelligent/unkind] × 3 [statement type: complete, omission, inaccurate] interaction, F[5.4, 207.9] = 47.17, p < .001, η2p = .55 [see Figure 1a].2 [Significant main effects also emerged for target, F[3, 115] = 43.96, p < .001, η2p = .53, and statement type, F[1.8, 203.7] = 294.43, p < .001, η2p = .72.] Testing the simple effect of statement type for each target separately showed that the likelihood of complete, omission, and inaccurate statements differed significantly for intelligent/kind, F[2, 58] = 295.84, p < .001, η2p = .91, intelligent/unkind, F[2, 62] = 110.77, p < .001, η2p = .78, and unintelligent/kind, F[1.6, 46.8] = 84.70, p < .001, η2p = .75, targets, but not the unintelligent/unkind target, F[2, 54] < 1, η2p = .03 [Table 1 reports means]. Because all statements were equally unlikely for the unintelligent/unkind target, this target was dropped from further analyses. Bonferroni-corrected pairwise comparisons showed that omission was more likely than complete accuracy in both ambivalent conditions [ps < .001], but not the univalent intelligent/kind condition [p = .103].3 Supplemental analyses confirmed that participants describing ambivalent targets tended to omit negative rather than positive content: “smart” exceeded “mean” [M = 3.13, SD = 1.16] for the intelligent/unkind target, F[1, 31] = 59.39, p < .001, η2p = .66, and “nice” exceeded “stupid” [M = 1.83, SD = 1.37] for the unintelligent/kind target, F[1, 29] = 113.93, p < .001, η2p = .80.

Likelihood of statements made to a casual acquaintance by vignette in Studies 1–3, respectively displayed in panels [a], [b], and [c]. Error bars indicate ± 1 SE.

Table 1

Statement-based Omission by Vignette and Audience in Studies 1–3

Discussion

Consistent with the negativity-omission hypothesis, participants opted for omission over complete accuracy in describing both types of ambivalent targets, but neither univalent target. Omission was systematic: Participants preferred to omit negative and emphasize positive dimensions of ambivalent targets’ behavior. Interestingly, “nice” exceeded “nice but stupid” [competence omission] to a greater extent than “smart” exceeded “smart but mean” [warmth omission], an asymmetry [evident across Studies 1–3] we explain in the Discussion. Following the quality maxim, participants mostly made accurate rather than inaccurate statements. When only negative information was available, such that all true statements violate the anti-negativity norm, participants did not opt for inaccurate positive statements over complete or omission statements. This null effect is inconsistent with a generic positivity bias, which would lead participants to describe a target favorably even if doing so contradicted factual information.

Study 2: Negativity Omission Across Audiences

We theorize that negativity omission arises from strategic self-presentation. Study 2 modifies the Study 1 procedure to influence self-presentation concerns. We manipulated the audience, theorizing that self-presentation concerns increase when addressing more public, unfamiliar audiences of casual acquaintances—who presumably have less elaborated or stable impressions of the speaker—as opposed to a familiar close friend or oneself privately. Work on TAR effects suggests that “in first encounters with listeners, it may behoove speakers to avoid professing overly negative impressions of third parties” [Ames et al., 2010, p. 158], but once audiences possess prior information about communicators—as close friends do—such information affects impressions of communicators more than their descriptions of others do [Gawronski & Walther, 2008]. Private audiences should reduce self-presentation concerns, though some concerns may linger due to evaluation apprehension [Rosenberg, 1969] about researchers’ impressions, or insofar as private thought prepares people for public performance via preemptive self-criticism, shaping underlying cognitive processes [Lerner & Tetlock, 1999].

This study also specifies an outgroup target from a historically stigmatized racial group, anticipating the group-level stereotyping-by-omission processes examined in Studies 4 and 5. Describing a Black target should heighten self-presentation concerns due to strong anti-racism norms [Crandall et al., 2002; Dunton & Fazio, 1997; Plant & Devine, 1998]; Whites who make negative remarks about a Black target elicit derogation and hostility [Simon & Greenberg, 1996].

Predictions

We expected participants to emphasize positive content and omit negative content when they receive ambivalent [not univalent] behavioral information about a target, particularly in public settings, although omission might also occur to some extent even in private.

Method

Participants

We recruited 125 undergraduates to complete an online study for a chance to win $10. Analyses excluded 8 prior Study 1 participants, 17 who failed manipulation checks, and 4 who identified as Black or African American [yielding an ingroup target]. The final sample [N = 96] comprised 66 women and 30 men, with a mean age of 20.6, including 66 self-identified Whites and 30 non-Black minorities. [Participant race did not qualify any results.]

Procedure and materials

Study 2 replicated Study 1, with two changes. First, the vignette specified target race: “Imagine someone named Pat, a black student of your same age, class year, and gender.” As before, the targets’ behavior was intelligent/kind [n = 16], intelligent/unkind [n = 28], unintelligent/kind [n = 28], or unintelligent/unkind [n = 24].4 Second, participants rated the likelihood of making each Study 1 statement “to yourself privately,” “to a close friend,” and “to a casual acquaintance.” Statements appeared on three randomly ordered pages [one per audience] in a within-participants design. Manipulation checks tested memory for target’s race and behavior, followed by demographic questions.

Results

Negativity omission

We first tested whether statement ratings [classified as in Study 1] from the casual acquaintance condition replicated Study 1 findings. The 3 [statement type: complete, omission, inaccurate] × 4 [target: intelligent/kind, intelligent/unkind, unintelligent/kind, unintelligent/unkind] interaction was significant, F[5.4, 165.9] = 43.78, p < .001, η2p = .59 [see Figure 1b]. [Significant main effects also emerged for statement type, F[1.8, 165.9] = 258.88, p < .001, η2p = .74, and target, F[3, 92] = 42.85, p < .001, η2p = .58.] As in Study 1, the likelihood of making complete, omission, and inaccurate statements to a casual acquaintance differed significantly for the intelligent/kind, F[1.3, 18.8] = 119.44, p < .001, η2p = .89, intelligent/unkind, F[1.4, 37.1] = 78.98, p < .001, η2p = .75, and unintelligent/kind, F[2, 54] = 159.45, p < .001, η2p = .86, targets, but not the unintelligent/unkind target, F[1.6, 36.0] = 3.05, p = .071, η2p = .11, so this last condition was dropped from further analysis [Table 1 reports means]. Bonferroni-corrected pairwise comparisons showed that omission was more likely than complete accuracy for the two ambivalent targets [ps < .001], but not the intelligent/kind target [p = .064].5 Participants describing ambivalent targets omitted negative more than positive content: “smart” trumped “mean” [M = 2.36, SD = 1.22] for the intelligent/unkind target, F[1, 27] = 82.57, p < .001, η2p = .67, and “nice” trumped “stupid” [M = 1.57, SD = 1.00] for the unintelligent/kind target, F[1, 27] = 207.44, p < .001, η2p = .89.

Omission across audiences

Next, we tested whether preference for omission over complete accuracy diminished when addressing more private audiences [i.e., a close friend or oneself]. Reasoning that self-presentation concerns rise monotonically when addressing oneself, a close friend, or a casual acquaintance, we computed linear-trend contrasts for audience publicity [weights: self = −1, friend = 0, acquaintance = +1]. We submitted these contrast values to a 3 [statement type: complete, omission, inaccurate] × 4 [target] ANOVA with statement type as a repeated measure. A significant linear trend of audience publicity emerged, F[1, 92] = 50.42, p < .001, η2p = .35, and as expected, this trend varied across different types of statements and targets, F[5.4, 164.9] = 7.53, p < .001, η2p = .20 [see Figure 2a]. A quadratic trend of audience emerged, F[1, 92] = 4.19, p = .044, η2p = .05, but did not vary by statement type and target, F[5.5, 167.5] = 1.23, p = .294, η2p = .04, so no further quadratic effects were tested.

Omission by audience and vignette. Panels [a] and [b] respectively display statement-based omission in Studies 2 and 3, and panel [c] displays open-ended omission in Study 3. Error bars indicate ± 1 SE. Acq. = acquaintance.

The negativity-omission hypothesis predicts variation across audiences in the likelihood of omission versus accuracy, but not inaccuracy. For inaccurate statements, the audience-publicity linear trend was neither significant, F[1, 92] = 1.66, p = .201, η2p = .02, nor variable across targets, F[3, 92] < 1, η2 = .01. In contrast, the likelihood of complete versus omission statements interacted significantly with target and audience-publicity linear trend, F[3, 92] = 7.66, p < .001, η2p = .20 [Table 1 reports means]. Participants’ relative preference to avoid complete accuracy in favor of omission increased significantly with more public audiences, but only for descriptions of ambivalent targets—intelligent/unkind, F[1, 27] = 18.71, p < .001, η2p = .41; unintelligent/kind F[1, 27] = 22.00, p < .001, η2p = .45—and not univalent targets—intelligent/kind, F[1, 15] < 1, η2p = .03, unintelligent/unkind, F[1, 23] < 1, η2p = .01. Thus, negativity omission prevailed over complete accuracy in descriptions of ambivalent targets made to more public audiences [see Figure 2a].

Supplemental trend analyses confirmed that participants were increasingly likely to omit negative rather than positive content when publicly describing ambivalent targets. Addressing a casual acquaintance versus oneself amplified the extent to which “smart” eclipsed “mean” for the intelligent/unkind target and “nice” eclipsed “stupid” for the unintelligent/kind target, respective Fs[1, 27] = 20.86 and 18.02, ps < .001, and η2ps = .44 and .40.

Target race

Although analyses pooling data across separate studies must be interpreted cautiously, we conducted a limited set of tests examining whether the previously documented omission effects for a target displaying ambivalent behavior are more pronounced for Black [Study 2] than race-unspecified [Study 1] targets. When addressing a casual acquaintance, the likelihood of making accurate versus omission statements varied marginally by target race for ambivalent [i.e., unintelligent/kind or intelligent/unkind] targets, F[1, 116] = 3.37, p = .069, η2p = .03, but not univalent targets, F[1, 96] < 1, η2p < .01. Omission [i.e., positive] statements were equally likely for Black versus race-unspecified ambivalent targets, F[1, 116] < 1, η2p < .01, but participants were less likely to make completely accurate [i.e., partially negative] statements about Black than race-unspecified targets, F[1, 116] = 9.20, p = .003, η2p = .07.

Discussion

Study 2 replicates the omission effects of Study 1 and extends them by showing that omission effects are larger for public than private audiences. As in Study 1, participants preferred omission to complete accuracy in describing ambivalent [not univalent] targets to an acquaintance, and the most likely statements about ambivalent targets were those omitting negativity and emphasizing positivity. Moreover, in describing ambivalent but not univalent targets, participants’ preference for avoiding completely accurate [partially negative] statements in favor of omission increased as the audience became more public. Additionally, participants describing an outgroup individual [a task presumably eliciting more self-presentation concerns] displayed a stronger negativity-omission pattern. These results suggest that people trade off complete accuracy for negativity omission in situations that heighten self-presentation concerns.

Study 3: Open-Ended Negativity Omission and Self-Presentation

We theorize that communicators omit negative information when describing ambivalent targets due to strategic self-presentation, not decreased concern about honesty or decreased certainty about ambivalent targets’ traits. We conceptualize strategic self-presentation as efforts to make socially appropriate statements to avoid the negative interpersonal consequences of publicly derogating others, such as appearing gossipy, judgmental, or impolite.

Study 2 provided evidence consistent with this self-presentation account: Participants increasingly opted for omission over complete accuracy when communicating with more public audiences, especially for an outgroup target. Study 3 affords more direct evidence of process by testing the extent to which self-presentation [vs. honesty or uncertainty] concerns mediate the effect of audience on preference for omission over complete accuracy. Study 3 also presents more conservative tests of omission by manipulating audience between [rather than within] participants and examining negativity omission in participants’ open-ended descriptions.

Predictions

We expected individuals to selectively emphasize positive content and omit negative content when given ambivalent [versus univalent] information about a target, especially with more public audiences. This effect should be mediated by participants’ self-presentation concerns, not honesty or uncertainty concerns. Moreover, omission should covary with self-presentation concerns only when describing ambivalent targets. Selectively stressing positive and omitting negative dimensions of ambivalent targets’ behavior lets communicators avoid violating anti-negativity norms when self-presentation matters. For univalent targets, however, evaluations have constant valence across dimensions, so anti-negativity norms should not lead to emphasizing one dimension over the other as a function of self-presentation.

Method

Participants

To earn course credit, 266 undergraduates completed an online study. Excluding 59 who failed manipulation checks, the sample [N = 207] comprised 8 who omitted demographic information, plus 122 women and 77 men with a mean age of 19.4, including 116 self-identified Whites and 83 minorities.

Procedure and materials

We randomly assigned participants to read a vignette from Study 1 about a race-unspecified target displaying intelligent/kind [n = 58], intelligent/unkind [n = 56], unintelligent/kind [n = 40], or unintelligent/unkind [n = 53] behavior. Study 3 used a between-participants audience manipulation, refined to distinguish more clearly the acquaintance and friend conditions in terms of self-presentation. We reasoned that talking to several unfamiliar acquaintances as opposed to one well-known friend elevates self-presentation concerns. We asked participants to “imagine that you are describing Pat [who is not present]” to one of the following randomly-assigned audiences: [a] “a couple casual acquaintances, whom you recently met and don’t know well,” [b] “a close friend, whom you befriended awhile ago and know very well,” or [c] “yourself in private.” To make the audience manipulation more salient, below this instruction a stick-figure schematic depicted a person talking to three others, talking to one other, or thinking to him- or herself, respectively.

We assessed omission using open-ended and statement-based measures. First, below the audience schematic, participants were asked to “write 2–3 sentences to describe Pat.” On the next page, participants rated their likelihood of making eight statements about the target from 1 [not at all likely] to 6 [extremely likely]. The Study 1 and 2 statements were modified such that “and” instead of “but” joined even opposite-valence descriptions [e.g., “Pat’s nice and stupid”], to rule out the possibility that “but” might connote ambivalence. Next, we had participants rate the extent to which 15 items tapping self-presentation, honesty, and uncertainty concerns [see Table 2] influenced their descriptions on a scale from 1 [not at all] to 7 [very much]. Manipulation checks probed memory for target behavior and audience, followed by demographic questions.

Table 2

Reliability and Principal Component Item Loadings of Influences on Descriptions in Study 3

Results

Statement-based omission

We tested whether participants’ likelihood of making complete, omission, and inaccurate statements to casual acquaintances paralleled our prior findings. A 3 [statement type: complete, omission, inaccurate] × 4 [target: intelligent/kind, intelligent/unkind, unintelligent/kind, unintelligent/unkind] ANOVA confirmed that willingness to make these statements to casual acquaintances varied across targets, F[5.3, 115.5] = 14.27, p < .001, η2p = .39 [see Figure 1c]. [Significant main effects emerged for statement type, F[1.8, 115.5] = 183.38, p < .001, η2p = .74, and target, F[3, 66] = 5.66, p = .002, η2p = .21.] The likelihood of making complete, omission, and inaccurate statements to a casual acquaintance differed for all targets: intelligent/kind, F[2, 34] = 113.30, p < .001, η2p = .87, intelligent/unkind, F[2, 38] = 61.91, p < .001, η2p = .77, unintelligent/kind, F[2, 22] = 55.47, p < .001, η2p = .84, and unintelligent/unkind, F[1.3, 25.5] = 10.31, p = .002, η2p = .35 [Table 1 reports means]. Bonferroni-corrected pairwise comparisons showed that omission was more likely than complete accuracy for both ambivalent targets [ps < .01], but neither univalent target [ps > .50].6 Participants omitted negative more than positive content in descriptions of ambivalent targets: “smart” trumped “mean” [M = 3.20, SD = 1.64] for the intelligent/unkind target, F[1, 19] = 25.50, p < .001, η2p = .57, and “nice” trumped “stupid” [M = 2.17, SD = 1.03] for the unintelligent/kind target, F[1, 11] = 106.48, p < .001, η2p = .91.

Statement-based omission between audiences

Next, we tested whether the tendency for omission to trump accuracy diminished for private audiences. Analyzing ratings of complete, omission, and inaccurate statements in a multilevel model revealed a significant audience- publicity linear trend, F[1, 216.0] = 9.04, p = .003, η2p = .04, qualified by a significant three-way interaction with target and statement type, F[6, 243.7] = 3.01, p =.007, η2p = .07 [see Figure 2b].7 [The main effects of statement type, F[2, 282.8] – 1143.49, η2p = .89, and target, F[3, 216.0] = 15.27, η2p = .17, both ps < .001, plus interactions of statement type with audience [linear trend], F[2, 243.7] = 6.10, p = .003, η2p = .05, and with target, F[6, 282.8] = 29.96, p < .001, η2p = .39, were also significant.] A non-significant quadratic trend, F[1, 216.0] < 1, η2p < .01, did not vary by target or statement type, F[6, 243.7] < 1, η2p = .02, so no further quadratic effects were tested.

The negativity-omission hypothesis does not predict variation in inaccuracy as a function of self-presentation concerns: For inaccurate statements, the audience-publicity linear trend did not vary by target, F[3, 195] < 1, η2p = .01. The next model excluded inaccuracy, revealing that the audience-publicity linear trend was qualified by a significant 4 [target] × 2 [statement type: accuracy, omission] interaction, F[3, 195] = 4.86, p = .003, η2p = .07 [Table 1 reports means]. Participants’ relative preference for omission over complete accuracy increased significantly with more public audiences, but only for descriptions of ambivalent targets—intelligent/unkind, F[1, 53] = 4.62, p =.036, η2p = .08; unintelligent/kind F[1, 37] = 10.55, p = .002, η2p = .22—not univalent targets—intelligent/kind, F[1, 55] .15 for all others]. As the audience became more public, target descriptions increasingly omitted warmth [p < .001, η2p = .25] but not competence [p = .747, η2p < .01] information about the intelligent/unkind target, and they increasingly omitted competence [p < .001, η2p = .29] but not warmth [p = .892, η2p < .01] information about the unintelligent/kind target [see Figure 2c].

Role of self-presentation concerns

To understand why participants increasingly omit negative content from descriptions of ambivalent targets when addressing more public audiences, we examined participants’ ratings of whether specific factors influenced their target descriptions [see Table 2]. A varimax-rotated principal components analysis of these items yielded three factors with eigenvalues over 1: self-presentation concerns, honesty concerns, and uncertainty concerns [Table 3 reports means by condition]. A three-factor model fit the data better than two- or one-factor models in confirmatory factor analysis, minimum Δχ2[1, N = 207] = 75.99, p < .001, indicating that the items indeed tapped three distinct concerns. Self-presentation concerns monotonically increased with audience publicity, F[2, 195] = 18.26, p < .001, η2p = .16, with significant differences between audiences: oneself [M = 3.06], friend [M = 3.85], acquaintances [M = 4.44], all ps < .05 [Bonferroni-corrected]. Honesty concerns also varied, to a lesser extent, across audiences, F[2, 195] = 5.36, p = .005, η2p = .05: Bonferroni-corrected comparisons confirmed lower honesty concerns for acquaintances [M = 4.19] than a friend [M = 4.82], p = .006, or oneself [M = 4.68], p = .049, while the latter two did not differ, p > .99. Uncertainty concerns did not vary by audience, F[2, 195] < 1, η2p < .01.

For both open-ended and statement-based omission, we tested potential mechanisms—self-presentation, honesty, and uncertainty concerns—in a multiple-mediator model [Preacher & Hayes, 2008], with the audience publicity linear trend [self = −1, friend = 0, acquaintance = +1] as the independent variable. The dependent variable was preference for omission over completely accurate statements [a difference score] in the statement-based omission model; in the open-ended omission model, it was warmth omission for the intelligent/unkind target and competence omission for the unintelligent/kind target. Because audience did not affect omission for univalent targets the mediation analyses necessarily focused on ambivalent targets [n = 96]. As predicted, self-presentation concerns significantly mediated the audience publicity effect on both statement-based and open-ended omission, respective Sobel zs = 3.81 and 4.00, ps < .001, whereas honesty and uncertainty concerns did not, zs < 1.35, ps > .15. The final model thus retained only self-presentation concerns as a mediator [see Figure 3]. Self-presentation concerns mediated the audience publicity effect on statement-based and open-ended omission for both intelligent/unkind targets, zs > 2.50, ps < .02, and unintelligent/kind targets, zs > 2.25, ps < .03.

Path analysis with self-presentation mediating the linear effect of audience publicity on statement-based [SB] omission versus accuracy and open-ended [OE] omission for participants describing ambivalent targets in Study 3. Coefficients are standardized regression betas.

***p < .001

Next, we tested whether self-presentation concerns predicted omission when describing ambivalent, not univalent, targets. [For univalent targets, open-ended omission was the mean of warmth and competence omission.] Target valence [univalent = −1; ambivalent = 1] moderated the relationship between self-presentation concerns and preference for statement-based omission over complete accuracy, β = .32, t[202] = 5.61, p < .001, as well as open-ended omission, β = .29, t[202] = 4.89, p < .001. Self-presentation concerns correlated with statement-based and open-ended omission strongly for intelligent/unkind and unintelligent/kind targets, all rs > .50, ps < .001, but not intelligent/kind and unintelligent/unkind targets, all rs < .15, ps > .30.

Discussion

These results replicate Studies 1 and 2: [a] participants opted for omission over complete accuracy in describing both ambivalent targets but neither univalent target; [b] ambivalent target descriptions systematically omitted negative rather than positive content; and [c] addressing more public audiences increased preference for omission over complete accuracy. Once again, descriptions of univalent negative targets were no more likely to include positive [inaccurate] than negative content. Extending these studies, Study 3 provides an additional measure of omission and evidence of process, while manipulating audience between [not within] participants. Like rating data, open-ended descriptions of ambivalent [not univalent] targets increasingly omitted negative [not positive] content when addressing more public audiences. As predicted, self-presentation concerns—not honesty or uncertainty concerns—mediated the effect of an increasingly public audience in producing more statement-based and open-ended omission. Self-presentation concerns were closely linked to omission for ambivalent but not univalent targets. In sum, Studies 1–3 support our claim that communicators, driven by self-presentation concerns, strategically omit negativity when they have ambivalent information about individuals.

Study 4: Stereotyping by Omission in the Updated Princeton Trilogy

The final two studies turn from descriptions of individual targets to stereotypes of groups. To test our group-level hypothesis about stereotyping by omission, we investigated omission of negative warmth and competence content in Princeton undergraduates’ reported stereotypes of 10 ethnic and national groups from the Princeton Trilogy stereotyping studies conducted in 1932, 1950, and 1967, as well as a modern sample. Using this archival data, we examined negativity omission based on reported positivity and negativity in groups’ historical stereotypes, many of which contained ambivalent content with markedly negative stereotypic warmth or competence. Notably, omitted dimensions of negative stereotypes are not assumed to be accurate; instead, parallel to negativity omission for individual targets, we assert that stereotyping by omission reflects incomplete disclosure of stereotypes that are “in the air” and broadly known to communicators regardless of personal prejudice levels [Devine, 1989].

Empirical and theoretical considerations discouraged us from attempting an audience publicity manipulation for communicators describing ethnic and national group targets. Publicity manipulations previously elicited discrepant levels of reported prejudice or discrimination [e.g., Carver, Glass, & Katz, 1978], but more recent data [e.g., Barker, 1994; Monin & Miller, 2001] show consistent judgments about outgroups across public and private audiences. Theoretically, internalizing anti-prejudice norms may motivate people to appear non-prejudiced not only publicly but also privately, agreeing that “I don’t want to appear racist or sexist, even to myself” or “I get angry with myself when I have a thought or feeling that might be considered prejudiced” [Crandall et al., 2002; Dunton & Fazio, 1997]. Rather than manipulate publicity, we examine omission processes through the lens of the Princeton Trilogy stereotyping studies conducted in 1932 [Katz & Braly, 1933], 1950 [Gilbert, 1951], 1967 [Karlins et al., 1969], and 2000–2007, four eras characterized by different norms regarding stereotyping.

Because the past eight decades have witnessed a steady increase in anti-prejudice norms, we expect that individuals have grown more likely to omit negative stereotype content over time. In the 1930s, prejudice against various ethnic and national groups was socially prescribed: many individuals feared losing status if they expressed positive attitudes toward stigmatized groups [Katz & F. H. Allport, 1931], and students reported more favorable outgroup stereotypes privately than publicly [Katz & Braly, 1935]. In the 1950s, the norms surrounding ethnic and racial stereotypes had become more mixed: G. W. Allport [1954/1979] observed that intolerant and tolerant norms varied across contexts, such that “anti-attitudes alternate with pro-attitudes. Often the see-saw and zig-zag are almost painful to follow” [p. 326]. In the late 1960s, anti-prejudice norms were strengthening: “Overall, the verbal norms in 1967 more nearly approach a vocabulary for friendly attitudes—a ‘language of tolerance’” [Karlins et al., 1969, p. 13]. Today, social norms strongly condemn expression of ethnic and national prejudices [Crandall et al., 2002], and many individuals report motivation to avoid appearing prejudiced [Dunton & Fazio, 1997; Plant & Devine, 1998], reflecting a reversal of the 1930s normative climate.

Predictions

Given contemporary anti-prejudice norms, we predict that individuals will omit historically negative dimensions of outgroup stereotypes while expressing positive ones. For example, an outgroup stereotyped as nice but stupid in the 1930s would not be described as intelligent today, but instead as possessing neutral or positive traits unrelated to competence. We did not anticipate stereotype reversal [i.e., negative stereotype dimensions becoming positive], uniform favorability increases across all groups and dimensions, or regression to the mean [i.e., refusal to stereotype, such that stereotype content appears diffuse and thus neutral]. Instead, we predicted a systematic shift in emphasis: retaining positive dimensions and ignoring negative ones, such that over time historically negative stereotype dimensions shift from expressed to omitted, while historically neutral or positive stereotype dimensions remain constant.

Our stereotyping-by-omission hypothesis leads to corollary predictions for stereotype continuity and favorability. If outgroup stereotypes converge toward neutrality on one [negative] dimension but not the other [positive] dimension, overall stereotype content should show only moderate continuity over time, with groups that were initially stereotyped negatively showing the least continuity. Historically negative outgroup stereotypes should increase in favorability [i.e., valence] over time as negative dimensions are dropped from mention, while positive stereotypes remain constant [not regressing toward the mean or showing a positivity boost].

Because historical stereotypes of more distant outgroups were especially negative, we expected them to improve more over time, relative to groups considered more similar to the ingroup. Thus, stereotypes of non-European outgroups should increase in favorability more than those of European outgroups, given earlier researchers’ observations that the most favorably viewed groups [besides Americans] consisted “exclusively of northern and western Europeans” [Katz & Braly, 1935, p. 176] in early stereotyping studies [Bogardus, 1925; Guilford, 1931; Thurstone, 1928] and that in 1950 the “tendency to favor Europeans [‘like us’] over non-Europeans [‘not like us’] [was] strong” [Karlins et al., 1969, p. 11]. By 1967, stereotypes of non-European groups had begun to improve, and we expected this trend to continue.

Our final prediction addresses participant demographic characteristics, which could provide an alternate—and arguably less interesting—account for the hypothesized changes in stereotype content. The 1932, 1950, and 1967 samples were exclusively male and predominantly White, but in 2000–2007 Princeton undergraduates were roughly 50% male and 65% White. If non-White or female students held more positive outgroup stereotypes, this demographic shift might explain stereotype favorability increases relative to prior White male samples. We predicted, however, that limiting the modern sample to White men for continuity with the 1932–1967 samples will yield patterns of stereotype favorability change identical to the full sample.

Other investigators have partially replicated the Katz-Braly methodology [e.g., Devine & Elliot, 1995; Dovidio & Gaertner, 1986; Lewis, Darley, & Glucksberg, 1972; Madon et al., 2001], but to our knowledge none have exactly replicated the full 10-group design and reported analyses for the demographic subgroup [White men] needed for historical continuity. Our study also updates the historical record by returning to the original Princeton participant population.

Method

Participants

We collected data from 135 Princeton undergraduates who participated for a psychology course or payment in 2000 [n = 75] or 2007 [n = 60]. The 2000 [and 2007] waves respectively comprised 43 [36] women and 32 [24] men, including 55 [37] identified as White and 20 [23] as non-White, with a mean age of 19.7 [19.8].8 The full sample included 28 freshmen, 57 sophomores, 22 juniors, 27 seniors, and 1 of unspecified year.

Procedure and materials

In the fall of 2000, participants completed a paper-and-pencil questionnaire either in class or at a psychology questionnaire day for $5. In the summer and fall of 2007, students completed the questionnaire online for either course credit or a chance to win $10. Replicating Katz and Braly [1933], participants selected from a list of 84 adjectives [in the original order] those that to them “seemed typical” of each of 10 ethnic or national groups [in counterbalanced order, with “African Americans” replacing “Negroes”], then identified the 5 most typical adjectives for each group. Finally, they reported demographic information.

Results

Stereotype content

All analyses focus exclusively on the top 5 adjectives selected by each participant as “most typical” for each group. We computed the percentage of participants who selected each adjective as “most typical” for each group [see Appendix A]. Analytic and theoretical factors determined the number and weighting of adjectives included in groups’ stereotypes. For non-parametric tests of continuity, stereotype content reflected whether each of 84 adjectives was [1] or was not [0] one of the 10 most frequently selected to describe the group. All other stereotype content indices were weighted by the percent of participants who selected an adjective as typical for a group. For stereotype favorability, stereotype content was based on the smallest number of adjectives needed to account for half of all possible responses [e.g., 338 = 5 adjectives × 135 participants ÷ 2] for each group [Appendix B reports the uniformity scores].

This definition of stereotype content as the most popular 50% of responses dates to Katz and Braly [1933], but Gilbert [1951] reports only the top 5 adjectives, plus any selected by at least 20% of participants, for each group. Karlins et al. [1969] retain the original stereotype content definition for their favorability score calculations, acknowledging that the 1950 scores—as they are based on fewer than 50% of responses—may be “less reliable” than scores for the other samples. For tests of stereotyping by omission over time, however, assessing a consistent proportion of responses for each sample is paramount; otherwise, dimensions might be “expressed” or “omitted” due to uneven reporting of content. For warmth- and competence-omission analyses, we defined stereotype content for each group as the top 5 adjectives [not the top 10] plus any selected by at least 20% of participants. Adopting this more conservative standard based on Gilbert [1951] afforded maximal consistency for omission analyses.

Finally, because the 2000 and 2007 samples reported highly consistent stereotype content for the 10 groups, average r[82] = .85, p < .001, Cohen’s kappa = .69, p < .001, we pooled their data to attain a sample size [N = 135] comparable to the earlier studies [Ns = 100, 333, 150] and acceptable cell sizes for subsample comparisons [e.g., White men vs. others].

Stereotype continuity

We tested stereotype continuity, defined as agreement on the top 10 adjectives for each group across time, using Cohen’s kappa. Stereotype content in 2000–2007 agreed significantly with that from each prior study for all groups except Japanese [K = −.09 relative to 1950, shortly after World War II] and the two originally most negatively evaluated groups: African Americans and Turks [see Appendix B]. Stereotype content agreement between current and prior samples was moderate for European groups [mean K = .51] and significantly lower for non-European groups [mean K = .32], t[2] = 7.36, p < .018. Overall, the group stereotypes show fair-to-moderate agreement over time, allowing for change.

Stereotype favorability

To assess stereotype favorability, we collected ratings of each adjective’s favorability from −2 [very unfavorable] to +2 [very favorable], like Karlins et al. [1969]. The modern ratings [see Appendix A], averaged across 2000 and 2007, r[82] = .98, p < .001, correlate highly with the 1967 favorability ratings, r[82] = .95, p < .001. To compute stereotype favorability, we multiplied the favorability of each adjective in a group’s stereotype by the number of participants selecting that adjective as typical of that group, then summed the products and divided by the total number of responses. Like Karlins et al., we applied the 1967 favorability ratings to prior years’ stereotype content, which could overlook subtle shifts in adjectives’ favorability prior to 1967. That said, the groups’ desirability ordering in Katz and Braly [1935] correlates .88 with their 1932 favorability scores based on 1967 adjective ratings.

The stereotype favorability scores for each group [see Table 3] revealed distinct patterns of change over time for the American ingroup, European outgroups [English, Irish, Italians, Germans, Jews9], and non-European outgroups [Chinese, Japanese, Turks, African Americans]. A 3 [group type: Americans, Europeans, non-Europeans] × 1 [linear trend of year: 1932linear trend of year: 1950linear trend of year: 1967linear trend of year: 200310] mixed-factorial ANOVA confirmed that favorability ratings for the 10 groups [averaged across years] did not vary by group type, F[2, 7] = 1.45, p = .297, η2p = .29, or year, F[3, 21] < 1, η2p = .05, or show a significant linear increase, F[1, 7] < 1, η2p = .08, but that group type interacted significantly with the year linear trend, F[2, 7] = 22.65, p < .001, η2p = .87, indicating that the shift in stereotype favorability over time varied across groups.

Follow-up analyses tested favorability linear trends over time separately for each group type. [No higher-order trends approached significance, all Fs < 2, ps > .25.] The American stereotype, historically the most positive, plummeted, but lacking other ingroups we could not statistically test this decrease. The moderately positive European stereotypes did not change in favorability over time, F[1, 4] < 1, η2 = .05, staying above zero in all periods, all Fs[1, 4] > 6.50, ps < .063, η2ps > .62. Stereotypes of non-European outgroups, historically the most negative, grew more favorable over time, F[1, 3] = 38.30, p < .001, η2p = .93.

Demographic variation

In 2000–2007, the White male subsample and overall sample agreed about the top 10 adjectives selected for each group, mean K = .80. A race-by-gender ANOVA [with group as a repeated measure] found no significant participant race or gender effects on 2000–2007 outgroup stereotype favorability scores between White men [n = 42], White women [n = 50], non-White men [n = 14], and non-White women [n = 29], all ps > .13. Stereotype favorability trend analyses comparing the earlier [mostly White male] samples with only the White men in the modern sample replicate the whole-sample results. Group type significantly interacted with the linear trend of year, F[2, 7] = 18.75, p = .002, η2p = .84, with stereotype favorability decreasing for the ingroup, staying constant for Europeans, F[1, 4] < 1, and increasing for non-European outgroups, F[1, 3] = 34.70, p = .010, η2p = .92. White men’s responses closely matched those of the overall sample in 2000–2007, suggesting that changes in stereotype content over time are not attributable to participant demographic shifts.

Stereotypic warmth and competence

To assess stereotypic warmth and competence, we asked nine stereotyping experts to categorize each adjective as reflecting warmth, competence, or neither dimension. Expert agreement averaged 86% for warmth words and 87% for competence words; Classifications reflect majority judgments [see Appendix A].

To confirm these classifications and ensure that they reflect the respective constructs theorized to underlie warmth and competence—intentions toward others and effectiveness in enacting these goals [Fiske et al., 2002; Wojciszke, 2005]—84 Princeton undergraduate judges [40 women, 44 men; mean age = 19.7] evaluated the adjectives on each dimension. Judges indicated on separate [counterbalanced] pages whether each adjective signaled a group’s “intention to help versus harm other groups” and “ability or inability to enact its intentions.” These ratings prompted 3 adjective reclassifications. Sly and unreliable—identified by over two-thirds of undergraduates as reflecting both tendency to harm and respective ability or inability—were reclassified as both warmth- and competence-related. Two-thirds of these judges did not classify loyal to family ties as a warmth [or competence] word, so it was classified as neither [see Appendix A]. The mean favorability of the respective warmth and competence adjectives in each group’s stereotype yielded warmth and competence scores for each group [see Table 4].11 If no warmth or competence adjectives were selected, the group’s score on that dimension was zero.

Table 4

Favorability of Group Stereotypes over Time in Study 4

Stereotyping by omission

To test the stereotyping-by-omission prediction that initially negative dimensions of stereotypes would be omitted over time, we classified group stereotypes as historically warm or cold based on their respective positive or negative 1932 warmth scores. Linear trend analyses of groups’ stereotype warmth scores revealed different trajectories for historically cold versus warm groups, F[1, 8] = 9.84, p = .014, η2p = .55. Stereotype warmth did not change over time for historically warm groups [English, Americans, African Americans], F[1, 2] < 1, η2p = .14, but increased for historically cold groups [Turks, Jews, Chinese, Japanese, Germans, Italians, Irish], F[1, 6] = 23.35, p = .003, η2p = .80 [see Figure 4a]. The cold groups’ stereotype warmth in 2000–2007 did not differ from zero, F[1, 6] = 1.26, p = .253, η2p = .21.

Stereotype content trends over time with groups classified by 1932 stereotypic warmth and competence in Study 4. Panels [a] and [b] respectively display warmth and competence.

Notably, classifying the Italian stereotype as historically cold was somewhat tenuous: Their warmth score was slightly negative [−0.23] in 1932 but distinctly positive [0.44] by 1950, on par with the warmest 1950 stereotypes [0.46]. To ensure that an anomalous or unstable Italian warmth stereotype did not cause the cross-temporal warmth increase for initially cold groups, trend analyses excluding Italians confirmed the warmth classification by year interaction, F[1, 7] = 7.99, p = .026, η2p = .53, and linear increase for historically cold groups, F[1, 5] = 16.21, p = .010, η2p = .76. Excluding Italians, all historically cold groups’ stereotypic warmth in 2000–2007 were omitted or neutral [i.e., between 0 and 0.10], significantly lower on average [M = 0.03] than the historically warm groups, F[1, 7] = 7.63, p = .028, η2p = .52.

Similarly, we classified groups’ stereotypes as historically competent, neutral, or incompetent based on their respective positive, neutral, or negative competence scores in 1932. Linear trend analyses of groups’ stereotype competence scores revealed marginally different patterns of change over time for historically competent, neutral, and incompetent groups, F[2, 7] = 3.68, p = .081, η2p = .51. Stereotypic competence did not change over time for historically competent groups [Chinese, Jews, Germans, Japanese, English, Irish, Americans], F[1, 6] < 1, η2p = .11, stayed consistently omitted over time for the historically neutral group [Italians], and increased to neutral for historically incompetent groups [African Americans, Turks], F[1, 1] = 115.37, p = .059, η2p = .99 [see Figure 4b]. For each historically incompetent or neutral group, stereotypic competence in 2000–2007 was neutral [0.00], lower than for historically competent groups, F[1, 7] = 3.92, p = .088, η2p = .36 [t[6] = −3.89, p = .008, not assuming equal variances].

Discussion

This analysis of stereotype content across 75 years supports the stereotyping-by-omission hypothesis. Similar to Studies 1–3, participants tended not to report uniform positivity. Rather than rate all groups equally favorably on all dimensions, participants reported stereotypes that largely matched our hypotheses about systematic, selective stereotype change. Initially favorable stereotypes [of European groups] remained moderately constant and positive, while initially unfavorable stereotypes [of non-European groups] improved substantially and dropped negative dimensions. Consistent with predictions for warmth and competence omission over time, modern participants described groups historically seen as cold [Turks, Jews, Chinese, Japanese, Germans, Italians, Irish] as now neutral in stereotypic warmth and groups historically seen as incompetent [African Americans, Turks] as now neutral in stereotypic competence, on average.

These results contrast with other possibilities. Given “negativity reversal,” historically negative dimensions of outgroup stereotypes would reverse to become positive, but we observed neutral scores on modern stereotypic warmth and competence for groups historically viewed negatively on these dimensions. “Uniform positivity” would lead all groups to improve on each dimension, but we found that trends depended on group type for favorability, warmth, and competence, with initially negative groups approaching neutral and positive groups remaining constant. “Refusal to stereotype” [low estimates of stereotype consensus] would lead initially positive and negative groups to average toward neutral over time, yet we found no decrease in stereotype favorability, warmth, and competence for groups initially stereotyped positively. In sum, the observed data fit the stereotyping-by-omission hypothesis better than alternate models.

Retaining positive stereotype dimensions while omitting negative ones perpetuates ambivalent stereotypes, in which outgroups are stereotyped as either warm or competent, but not both. We ran Study 5 to confirm negativity omission and ambivalence in modern stereotypes.

Study 5: Negativity Omission in Modern Stereotype Content

Study 5 validates and extends the Study 4 by using more fine-grained measures of warmth and competence to assess modern stereotype content for the 10 Katz-Braly groups. Although the adjective favorability ratings, which date back to the 1960s and converge with desirability ranking from the 1930s, are particularly well suited to historical trend analyses, direct assessments of warmth and competence more stringently test the stereotyping-by-omission hypothesis and more closely examine modern ambivalent outgroup stereotypes.

We recruited two new samples of Princeton undergraduates to rate the groups’ warmth and competence using different methods. The first “Katz-Braly adjectives” method inverted the Study 4 design by providing adjectival descriptions—but no names—for each group and having participants more precisely quantify the groups’ warmth and competence. The second Stereotype Content Model [SCM] method uses the standard SCM measure [Fiske et al., 2002]: Participants rated each named group’s warmth and competence using standard SCM traits. We aimed to rule out a “displacement” account that explains omission of negative stereotypes as an artifact of the Katz-Braly [1933] adjective-checklist method: As selecting only 5 top adjectives per group is a zero-sum exercise, more salient dimensions might “displace” negative dimensions. For example, top German adjectives might reflect very positive stereotypic competence, not lack of warmth.

Predictions

We expected stereotype content findings to show high reliability with groups’ stereotypic warmth and competence scores from Study 4, with mostly positive and neutral, not negative, stereotypes reported. Outgroups’ positions in a warmth-by-competence stereotype-content matrix are predicted to fan out across the neutral-to-positive space.

In the Princeton Trilogy data, only two outgroups’ historical stereotypes were univalent with respect to warmth and competence: Stereotypes of the English as warm and competent [e.g., courteous, intelligent] and Turks as cold and incompetent [e.g., cruel, ignorant] were evident in 1932 and 1950. Other outgroups’ historical stereotypes reflected bidimensional ambivalence: warm-and-incompetent [e.g., African Americans] or cold-and-competent [e.g., Jews] content, occasionally with mixed-valence content even within a dimension [e.g., the pugnacious, quick-tempered, yet honest Irish 1932 warmth stereotype, or Italians’ negative 1932 and positive 1950 warmth]. We predicted that Study 5 rating methods would classify the English stereotype as high on both warmth and competence, retaining positivity, while classifying the Turks as neutral on both dimensions, omitting negativity but not adding positivity. Stereotypes of the other 7 historically ambivalent outgroups were expected to fall into ambivalent clusters with one selectively retained positive dimension and one omitted [historically negative] dimension.

Method

Participants

The first sample [N = 258] included 155 women and 100 men; 149 Whites and 106 non-Whites; 64 freshmen, 114 sophomores, 29 juniors, and 48 seniors; plus 3 omitting demographic information.12 The second sample [N = 165] comprised 94 women and 71 men; 102 Whites and 63 non-Whites; 53 freshmen, 43 sophomores, 36 juniors, 32 seniors, and 1 of unspecified year. Participants received course credit.

Procedure and materials

Both methods used a within-participants design: Participants rated all 10 Katz-Braly groups. The first sample rated the apparent warmth and competence of unnamed groups conveyed by the adjectives selected for each group in Study 4: Participants viewed a sample adjective checklist for a non-existent group, then rated the warmth and competence of the anonymous groups [in random order] based on adjectival descriptions. The second sample completed the SCM measure for the named groups in one of two orders.

For the Katz-Braly adjectives method, we asked the first sample to read then rate descriptions of specific groups, such as “national groups, ethnic groups, religious groups, or other types of groups.” Rating pages presented the most frequently selected 5 adjectives, plus any selected by at least 20% of the sample, to describe each group in Study 4, without identifying the group. For example, one page read: “This group was described by prior research participants as ‘passionate, loyal to family ties, talkative, loud, pleasure-loving, artistic, sensual, suave.’ Based on this description, to what extent do you think this group possesses the following characteristics?” Participants rated the extent to which each group appeared warm [friendly, cold [reversed], likable; α = .82] and competent [capable, incompetent [reversed], smart; α = .87] from 1 [not at all] to 7 [extremely]. For the SCM method, the second sample rated named groups on standard traits for warmth [warm, good-natured, sincere, friendly; α = .83] and competence [competent, intelligent, capable, skillful; α = .89] from 1 [not at all] to 5 [extremely].

Results

We calculated mean warmth and competence ratings—averaged across traits and participants—for each group [see Table 5]. Stereotypes of outgroups assessed using the Katz-Braly adjectives versus standard SCM traits correlated positively for warmth, r[7] = .95, p < .001, and competence, r[7] = .94, p < .001. Study 4 stereotype-content scores and both Study 5 outgroup ratings proved reliable for warmth [α = .82] and competence [α = .95]. Thus, despite the classic adjective checklist methodology’s limitations [see Devine & Elliot, 1995], its stereotype-content findings match modern assessments.

Table 5

Warmth and Competence of Group Stereotypes over Time in Study 4

Bidimensional ambivalence in outgroup stereotypes emerged in correlational and cluster analyses. Outgroups’ stereotypic warmth and competence tended to correlate negatively, whether based on Katz-Braly adjectives, r[7] = −.71, p = .034, or standard SCM traits, r[7] = −.50, p = .168, consistent with compensatory stereotyping [Fiske et al., 2002; Judd et al., 2005; Kervyn, Yzerbyt, Demoulin, & Judd, 2008].

We plotted the groups in a warmth-by-competence matrix based on ratings of stereotype content using each method. Hierarchical cluster analysis of groups’ stereotypic warmth and competence using Ward’s method with squared Euclidean distance yielded a four-cluster solution for each sample, and a K-means cluster analysis defined the four clusters [see Figure 5]. Groups fanned out across the neutral-to-positive portion of the warmth-by-competence space, with cluster centers located near or above 4 [range: 1 to 7] for ratings based on Katz-Braly adjectives and near or above 3 [range: 1 to 5] for ratings made with standard SCM traits. Cluster membership showed 90% agreement [100% for outgroups] across methods for the four clusters:

1. high warmth and competence [HW/HC]: English; plus Americans [SCM method]

2. neutral warmth/high competence [NW/HC]: Germans, Jews, Chinese, Japanese

3. high warmth/neutral competence [HW/NC]: Italians, Irish, African Americans

4. neutral warmth and competence [NW/NC]: Turks; plus Americans [adjective method]

Stereotype content assessed with two methods in Study 5. Panel [a] reports stereotype content based on Katz-Braly adjectives selected for each group in 2000–2007 [range = 1–7] and panel [b] displays stereotype content SCM Likert scale ratings for each group [range = 1–5].

Thus, this analysis placed the historically univalent groups [English, Turks] in the predicted clusters characterized by the uniform inclusion or omission of both warmth and competence. The other 7 historically ambivalent outgroup stereotypes were located in ambivalent clusters, whose centers [see Table 6] differed significantly on stereotypic warmth and competence for both assessment methods. Warmth exceeded competence in the HW/NC cluster, ts[2] > 8, ps < .02, and competence exceeded warmth in the NW/HC cluster, ts[3] > 9, ps < .01. Mean warmth and competence ratings did not differ in the HW/HC or NW/NC clusters, ts[1] < 2, ps > .25.

Table 6

Stereotype Content Cluster Centers and Group Means [with SDs] in Study 5

Discussion

As predicted, two methods of assessing the Katz-Braly groups’ contemporary stereotype content converge with Study 4 warmth and competence scores, confirming the adjective-checklist results and opposing a displacement account of stereotyping-by-omission findings. Study 5 also reveals pervasive bidimensional ambivalence in modern outgroup stereotype content, showing a compensatory relationship between stereotypic warmth and competence: Most outgroups fall into clusters characterized by positivity on one dimension and neutrality on the other. These results imply that omitting historically negative stereotype dimensions while retaining positive ones perpetuates systematic bidimensional ambivalence toward outgroups.

General Discussion

Five studies show that communicators describing ambivalent individuals and groups accentuate perceived positive characteristics and eliminate negative ones, avoiding statements that are accurate but negative or positive but inaccurate. For ambivalent [intelligent/unkind, unintelligent/kind] but not univalent [intelligent/kind, unintelligent/unkind] individual targets, communicators opted for negativity omission over complete accuracy [Studies 1–3]. Omission superseded complete accuracy most when describing ambivalent targets to a public audience as opposed to more private audiences [Studies 2 & 3], and self-presentation concerns mediated the effect of audience publicity on omission [Study 3].

Paralleling individual-level negativity omission, reported stereotypes of 10 ethnic and national groups over 75 years revealed that as anti-prejudice norms grew stronger, historically negative aspects of stereotypes faded from mention [Study 4]. Cold or incompetent group stereotypes from the 1930s did not reverse to become respectively warm and competent in the modern era, but instead became neutral, while positive stereotypes persisted. Modern stereotype assessment methods replicated the contemporary pattern of often-ambivalent stereotype content, with outgroups rated favorably on one dimension and neutrally on another [Study 5]. Negativity omission was if anything more pronounced for non-White targets [e.g., Black individuals or non-European groups] than for others [e.g., race-unspecified individuals or European, phenotypically White groups], in contexts that exert strong anti-negativity pressures [Studies 2 & 4].

As noted, negativity omission emerged asymmetrically by domain in Studies 1–3. Both closed- and open-ended measures indicated more competence omission for unintelligent/kind targets than warmth omission for intelligent/unkind targets. Likewise, only competence, not warmth, was omitted entirely from specific outgroup stereotypes in Study 4 prior to 2000. This asymmetry parallels warmth’s functional primacy over competence in social cognition. Warmth information “typically has a direct and unconditional bearing on the well-being of other people” [Wojciszke, 2005, p. 16], hence “other-profitable” [see Peeters, 1992]; competence information is “self-profitable” for the target, but affects others indirectly via target goals. The negativity-omission hypothesis does not require identical omission of negative content across domains.

Our analytic framework differentiates between positivity expression [saying favorable things, even if they are not true] and negativity omission [not saying unfavorable things, even if they are true], revealing asymmetries. Specifically, negativity omission occurs for more types of targets and trumps positivity expression when information about targets is ambivalent. Self-presentation and prejudice concerns had a larger impact on negativity omission than on positivity expression, and perceived target traits limited positivity expression more than negativity omission. When self-presentation was salient, participants hesitated to criticize all targets—regardless of whether individuals’ behaviors or groups’ stereotypes had negative content—but their expression of positivity was still constrained by [subjective] accuracy. Descriptions made to acquaintances were mostly positive for three targets, but not the fourth [unintelligent/unkind], whereas negativity levels were low for all four targets [Studies 1–3]. Likewise, positivity emerged in the modern sample for eight [ambivalent] outgroup stereotypes, but not the Turks, whereas negativity was absent for all outgroups [Studies 4 & 5]. Per the negativity-omission and stereotyping-by-omission hypotheses, eliminating negativity and accentuating positivity occurred for only ambivalent targets, but descriptions of univalent negative targets [unintelligent/unkind individuals, Turks] also displayed relatively little derogation. Participants hesitated to endorse any statements or traits for univalent negative targets, as evidenced by low ratings for all types of statements about unintelligent/unkind targets [Studies 1–3] and the uniquely high participant skip rate plus low consensus for the Turkish stereotype [Study 4]. Our findings fit the maxim that communicators with nothing nice to say should say nothing at all, not fake positivity: Better to violate accuracy norms with “sins” of omission than commission.

Limitations

Conversational context

Our work relies on simulated communication, as opposed to actual live conversations. Because self-presentation concerns increase omission, we predict that addressing live [vs. fictive] audiences further elevates self-presentation concerns, amplifying negativity omission, so our studies estimate omission effects conservatively. In the absence of any real relationship with their audience, participants in psychology studies still routinely respond as though audience approval matters, and live audiences are theorized to boost accountability further [Lerner & Tetlock, 1999], potentially increasing communicators’ desire to avoid responsibility for expressing negative content “on the record” [Brown & Levinson, 1978].

Ingroup evaluations

Stereotypes of the American ingroup varied across Studies 3 and 4, sometimes contrasting with the stereotyping-by-omission effects observed for outgroups. Of the 10 groups assessed, only the American stereotype grew markedly less favorable over time—dropping from most to least positive—and was inconsistently classified by the two Study 5 rating methods. These mixed ingroup characterizations do not contradict our omission hypotheses, which assert that communicators describing others omit negative content due to self-presentation concerns, because different norms govern self- and other-directed negativity. Criticism of groups from ingroup as opposed to outgroup sources is perceived more favorably [Hornsey, Oppes, & Svensson, 2002] and deemed more normatively appropriate [Sutton et al., 2006].

Princeton Trilogy constraints

Longitudinal analyses of Princeton Trilogy data do not permit testing whether strengthening anti-prejudice norms account for changes in expressed stereotypes. Nevertheless, our account converges with expert interpretations of Princeton Trilogy stereotyping data [e.g., Devine & Elliot, 1995; Karlins et al., 1969; Madon et al., 2001], and our analyses using only White male students preclude ascribing these changes to demographic shifts.

Admittedly, the Katz-Braly adjective-checklist method relies on 1930s adjectives and fails to differentiate personal and societal stereotypes [Devine & Elliot, 1995]. That said, all adjectives except stolid were familiar to over 90% of Princeton undergraduates in pilot testing, and Study 5 validates Study 4 results, confirming these adjectives’ ability to assess stereotype content. Because we focus on stereotype expression, and expect historically negative dimensions of both personal and societal stereotypes to be omitted, probing the relative influence of personal beliefs and societal representations on reported stereotypes is beyond the scope of our analysis.

Using Princeton undergraduates—“social descendants” of the original Princeton Trilogy samples—in our replication presents limitations and strengths. Students at an elite private university do not represent Americans more generally, yet because graduates of elite schools are overrepresented in societal leadership positions, shaping public policy and opinion, their responses may serve as bellwethers of broader societal shifts to come.

Finally, the Princeton Trilogy data provide more stereotype depth than breadth, including only 9 outgroups. [As outgroup analyses with n = 9 have low statistical power, even marginal findings reflect large effects, η2p > .30.] The 9 Katz-Braly groups span several continents and vary broadly in stereotypic warmth and competence, but many salient groups are absent. Studies of other groups do show effects consistent with stereotyping by omission: Outgroups toward whom prejudice is normatively prescribed [e.g., drug addicts, welfare recipients; Crandall et al., 2002] elicit decisively negative stereotypic warmth and competence ratings [Fiske et al., 2002], whereas outgroups toward whom prejudice is normative proscribed [e.g., elderly people, Hispanics, business women, Jews, “mentally retarded” people] occupy the neutral-to-positive part of a stereotypic warmth-by-competence matrix, with stereotypes located in ambivalent clusters defined by positivity on one dimension and neutrality [or slight negativity] on the other.

Implications and Future Directions

Two features of social cognition—its heavy weighting of negative information and collaborative nature—underscore the potentially far-reaching implications of negativity omission and stereotyping by omission. First, negative social information is valued because in theory less desirable behaviors prove more diagnostic for trait attributions [Jones & Davis, 1965] and in practice people show greater sensitivity to negative than positive social information [Fiske, 1980; Kanouse & Hanson, 1972], more rapidly forming and less readily disconfirming negative [vs. positive] impressions and stereotypes [Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001]. Social cognition is socially situated, distributive, and collaborative [Smith & Collins, 2009]. Secondhand information shapes impressions of targets and communicators [Ames, Bianchi, & Magee, 2010; Ames & Welch, 2011] spontaneously during conversation [Wyer et al., 1990]. Biases toward negativity in attention and positivity in expression [see Lewicka et al., 1992] create a dilemma: Audiences value receiving negative content about targets, but communicators more readily share positive information. Omitting negativity violates the quantity maxim of cooperative conversation, hindering communication.

At an intergroup level, omission of negative stereotypes sounds beneficial for society and stigmatized groups in particular, but also poses pitfalls. Discussing positive aspects of intergroup relations increases perceived intergroup harmony yet ironically preserves intergroup inequalities [Saguy, Tausch, Dovidio, & Pratto, 2009]. Omitting negativity may foster a non-prejudiced self-image that subjectively licenses communicators to “express prejudice in ambiguous ways, feigning rejection of the stereotype but still managing to express it” [Crandall & Eshleman, 2003, p. 423; see Monin & Miller, 2001]. Outwardly omitting negative stereotypes may foster illusory perceptions of racism persisting without racists or not existing at all.

Conversely, omitted negativity may not be lost, but instead inferred by audiences. We focus on communicators’ selective expression of information, but communicated impressions also depend on audiences’ reception of this filtered content. If only more readily communicable traits propagate over time [see Schaller et al., 2002], and negative content is selectively omitted, negative impressions of targets would eventually disappear altogether. Instead, we suggest that audiences may detect strategic omission, just as they decode negativity implied by outwardly positive descriptions [e.g., a “really nice” job applicant, a potential date with “a great personality”]. The term stereotyping by omission reflects primarily our claim that systematically omitting dimensions constitutes a form of stereotyping, but also our intuition that conspicuous omission implies negativity, covertly conveying negative aspects of stereotypes.

Emerging evidence of innuendo effects supports the claim that audiences draw negative inferences from communicators’ faint or unidimensional praise of targets. A recommendation letter from an expert praising a target’s politeness while omitting his competence lowered target evaluations [Harris, Corner, & Hahn, 2009]. When communicators omitted information on a salient dimension of social perception [e.g., describing a prospective travel partner as “very smart, hard-working, and competent” or a work partner as “very nice, sociable, and outgoing”] audiences drew negative inferences about the target on the omitted dimension and were more likely to exclude the target as a function of these inferences [Kervyn, Bergsieker, & Fiske, 2012]. These innuendo-based inferences persisted in a communication chain more readily for female [vs. male] targets in work contexts, consistent with warm-but-incompetent default stereotypes of women [Glick & Fiske, 1996] and the emphasis on warmth in recommendations for female [vs. male] academics, which negatively predicted hiring [Madera, Hebl, & Martin, 2009].

At an intergroup level, the claim that omitted negative dimensions of stereotypes may be inferred fits existing theorizing and receives indirect empirical support. The often compensatory relationship between outgroups’ inferred warmth and competence [Judd et al., 2005; Yzerbyt et al., 2008], and tight coupling of positive and negative stereotypes [Czopp & Monteith, 2006], mean that persistently emphasizing positive unidimensional outgroup stereotypes may hinder repair of closely linked [but unspoken] negative stereotypes on other dimensions. When White or male communicators respectively praise Blacks’ athletic ability or say “You did really well for a woman,” such allegedly positive “compliments” lead [Black or female] audiences to draw negative inferences about communicators’ impressions [Czopp, 2008; Garcia, Miller, Smith, & Mackie, 2006]. Similarly, White communicators give more positive open-ended descriptions of Black than White targets, but audiences who know the targets’ race interpret these descriptions more negatively, forming worse impressions of Black than White targets [Collins, Biernat, & Eidelman, 2009]. Future work should test whether hearing ostensibly positive group stereotypes expressed in one domain leads audiences to infer specific stereotypic deficiencies in others.

If historically negative outgroup stereotype dimensions become neutral but not favorable, and positivity in one area may lead to compensatory backsliding in another, a forecast for ethnic and national outgroup stereotypes 75 years from now would involve not continued favorability increases, with most outgroups seen as warm and competent, but instead stereotype stagnation. When societal norms strongly discourage expressions of prejudice, omission-based processes may subtly perpetuate negative representations of groups, allowing latent negative stereotype dimensions to rebound not only in the minds of communicators but also in the inferences of their audiences. Stereotyping by omission thus adds to the evidence that merely suppressing prejudice is unlikely to improve intergroup relations.

Acknowledgments

This work was supported by a National Science Foundation Graduate Research Fellowship to the first author under Grant No. DGE–0646086 and Princeton University funding. We thank Daniel L. Ames, Ido Liviatan, Nicolas Kervyn, Betsy Levy Paluck, J. Nicole Shelton, Stacey Sinclair, Deborah Son Holoien, and Valerie Jones Taylor for comments on this manuscript and Berenice Jimenez for research assistance.

Appendix A: Favorability Rating and Percentage of Participants Selecting Adjectives by Group in 2000–2007

Appendix B: Stereotype Uniformity Scores, Non-Participation, and Continuity Scores by Group over Time

Footnotes

Portions of Studies 4 and 5 were presented at the Society for Personality and Social Psychology annual conference in February 2009 and summarized in Fiske, S. T., Bergsieker, H. B., Russell, A. M., & Williams, L. [2009]. Images of Black Americans: Then, “them” and now, “Obama!” DuBois Review: Social Science Research on Race, 6, 83–101. The first wave of Study 4 data came from the second and third authors’ senior theses at Princeton University.

1Twenty students rated the valence of 30 warmth and competence descriptors on a 7-point scale. Smart and nice did not differ [Ms = 5.60 & 5.20], F[1, 19] = 3.55, p = .08, nor did stupid and mean [Ms = 2.35 & 2.15] F[1, 19] < 1.

2Fractional df reflect a Greenhouse-Geisser correction.

3Inaccuracy was less likely than [a] omission for the three remaining targets [ps < .001] and [b] complete accuracy for the intelligent/kind and intelligent/unkind targets [ps < .001] but not the unintelligent/kind target [p = .808].

4A programming error randomly assigned fewer participants to the intelligent/kind cell.

5Inaccuracy was less likely than [a] omission for the three remaining targets [ps < .001] and [b] complete accuracy for the intelligent/kind and intelligent/unkind targets [ps < .005], but not the unintelligent/kind target [p > .99].

6Inaccuracy was less likely than [a] omission for all targets [ps < .001] and [b] complete accuracy for both intelligent targets [ps < .001], but not unintelligent/kind [p = .240] or unintelligent/unkind [p = .059] targets.

7All tests of interactions between repeated measures and the linear or quadratic trends of audience [manipulated between participants] involved multilevel models run using the SPSS mixed procedure with a custom L matrix. Estimating heterogeneous error terms for repeated measures can result in fractional df.

8Age data are not available for the psychology course participants [mostly freshmen or sophomores] in 2000.

9Classifying Jews as non-European does not alter results. Jewish stereotypes straddle the European/non-European favorability divide in early work [e.g., Bogardus, 1925; Guilford, 1931; Thurstone, 1928]. The trajectory resembles both European stereotypes [moderately positive in 1932] and non-European stereotypes [increasing favorability], with a gain [+0.45] halfway between European [−0.05] and non-European [+0.84] groups.

10Custom polynomial SPSS syntax adjusted for unequal intervals; 2003 is the weighted average year for our sample.

11We inverted 2 adjectives’ favorability scores to appropriately capture their warmth- or competence-specific valence. Sly—defined by Merriam-Webster’s Collegiate Dictionary [2005] as “clever in concealing one’s aims or ends” thus had competence scores 0.58 and 0.72, respectively, for prior and modern samples, and stolid—defined as “having or expressing little or no sensibility; unemotional”—had warmth score of −0.32 [not 0.32] for prior samples.

12This sample did Study 2 previously within a set of online studies, separated by unrelated measures. No one noted any suspicion that the studies were related, and Study 2 manipulations did not affect Study 5 results, all Fs < 1.

Contributor Information

Hilary B. Bergsieker, Department of Psychology, Princeton University.

Lisa M. Leslie, Carlson School of Management, University of Minnesota.

Vanessa S. Constantine, Department of Psychology, Princeton University.

Susan T. Fiske, Department of Psychology, Princeton University.

References

• Allport GW. The nature of prejudice. Cambridge, MA: Perseus Books; 1954/1979. [Google Scholar]
• Ames DR, Bianchi EC, Magee JC. Professed impressions: What people say about others affects onlookers’ perceptions of speakers’ power and warmth. Journal of Experimental Social Psychology. 2010;46:152–158. [Google Scholar]
• Ames DR, Welch WT. The ecology of everyday first encounters: Prevalent circumstances and the impact of individual differences. 2011. Unpublished manuscript. [Google Scholar]
• Anderson NH. Likableness ratings of 555 personality-trait words. Journal of Personality and Social Psychology. 1968;9:272–279. [PubMed] [Google Scholar]
• Armitage CJ. Beyond attitudinal ambivalence: Effects of belief homogeneity on attitude-intention-behaviour relations. European Journal of Social Psychology. 2003;33:551–563. [Google Scholar]
• Backbier E, Hoogstraten J, Terwogt-Kouwenhoven K. Situational determinants of the acceptability of telling lies. Journal of Applied Social Psychology. 1997;27:1048–1062. [Google Scholar]
• Barker K. To be PC or not to be? A social psychological inquiry into political correctness. Journal of Social Behavior and Personality. 1994;9:271–281. [Google Scholar]
• Baumeister RF, Bratslavsky E, Finkenauer C, Vohs KD. Bad is stronger than good. Review of General Psychology. 2001;5:323–370. [Google Scholar]
• Baumeister RF, Leary MR. The need to belong: Desire for interpersonal attachments as a fundamental human motivation. Psychological Bulletin. 1995;117:497–529. [PubMed] [Google Scholar]
• Bergsieker HB, Monin B. Whom to hate: The role of norms and attributions in regulating prejudice. Poster presented at 7th Annual Conference of the Society for Personality and Social Psychology in Palm Springs; CA. 2006. [Google Scholar]
• Blanchard FA, Crandall CS, Brigham JC, Vaughn LA. Condemning and condoning racism: A social context approach to interracial settings. Journal of Applied Psychology. 1994;79:993–997. [Google Scholar]
• Bogardus ES. Measuring social distances. Journal of Applied Sociology. 1925;9:299–308. [Google Scholar]
• Bok S. Lying: Moral choice in public and private life. New York: Vintage; 1978. [Google Scholar]
• Boucher J, Osgood CE. The Pollyanna hypothesis. Journal of Verbal Learning and Verbal Behavior. 1969;8:1–8. [Google Scholar]
• Brown P, Levinson SC. Politeness: Some universals in language usage. Cambridge, UK: Cambridge University Press; 1987. [Google Scholar]
• Carver CS, Glass DC, Katz I. Favorable evaluations of blacks and the handicapped: Positive prejudice, unconscious denial, or social desirability? Journal of Applied Social Psychology. 1978;8:97–106. [Google Scholar]
• Cialdini RB, Richardson KD. Two indirect tactics of image management: Basking and blasting. Journal of Personality and Social Psychology. 1980;39:406–415. [Google Scholar]
• Collins EC, Biernat M, Eidelman S. Stereotypes in the communication and translation of person impressions. Journal of Experimental Social Psychology. 2009;45:368–374. [Google Scholar]
• Crandall CS, Eshleman A. A justification-suppression model of the expression and experience of prejudice. Psychological Bulletin. 2003;129:414–446. [PubMed] [Google Scholar]
• Crandall CS, Eshleman A, O’Brien LT. Social norms and the expression and suppression of prejudice: The struggle for internalization. Journal of Personality and Social Psychology. 2002;82:359–78. [PubMed] [Google Scholar]
• Cuddy AJC, Fiske ST, Glick P. The BIAS map: Behaviors from intergroup affect and stereotypes. Journal of Personality and Social Psychology. 2007;92:631–648. [PubMed] [Google Scholar]
• Czopp AM. When is a compliment not a compliment? Evaluating expressions of positive stereotypes. Journal of Experimental Social Psychology. 2008;44:413–420. [Google Scholar]
• Czopp AM, Monteith MJ. Thinking well of African Americans: Measuring complimentary stereotypes and negative prejudice. Basic and Applied Social Psychology. 2006;28:233–250. [Google Scholar]
• Devine PG. Stereotypes and prejudice: Their automatic and controlled components. Journal of Personality and Social Psychology. 1989;56:5–18. [Google Scholar]
• Devine PG, Elliot AJ. Are racial stereotypes really fading? The Princeton trilogy revisited. Personality and Social Psychology Bulletin. 1995;21:1139–1150. [Google Scholar]
• Dovidio JF, Gaertner SL. Prejudice, discrimination, and racism: Historical trends and contemporary approaches. In: Dovidio JF, Gaertner SL, editors. Prejudice, discrimination, and racism. Orlando, FL: Academic Press; 1986. pp. 1–34. [Google Scholar]
• Dunton BC, Fazio RH. An individual difference measure of motivation to control prejudiced reactions. Personality and Social Psychology Bulletin. 1997;23:316–326. [Google Scholar]
• Ekman P. Telling lies: Clues to deceit in the marketplace, politics, and marriage. New York, NY: Norton; 2001. [Google Scholar]
• Feather NT, Armstrong DJ. Effects of variation in source attitude, receiver attitude, and communication stand on reactions to source and content of communications. Journal of Personality. 1967;35:435–455. [PubMed] [Google Scholar]
• Fiske ST. Attention and weight in person perception: The impact of negative and extreme behavior. Journal of Personality and Social Psychology. 1980;38:889–906. [Google Scholar]
• Fiske ST, Cuddy AJC, Glick P. Universal dimensions of social perception: Warmth and competence. Trends in Cognitive Science. 2007;11:77–83. [PubMed] [Google Scholar]
• Fiske ST, Cuddy AJ, Glick P, Xu J. A model of [often mixed] stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality and Social Psychology. 2002;82:878–902. [PubMed] [Google Scholar]
• Folkes VS, Sears DO. Does everybody like a liker? Journal of Experimental Social Psychology. 1977;13:505–519. [Google Scholar]
• Garcia AL, Miller DA, Smith ER, Mackie DM. Thanks for the compliment? Emotional reactions to group-level versus individual-level compliments and insults. Group Processes and Intergroup Relations. 2006;9:307–324. [Google Scholar]
• Gawronski B, Walther E. The TAR effect: When the ones who dislike become the ones who are disliked. Personality and Social Psychology Bulletin. 2008;34:1276–1289. [PubMed] [Google Scholar]
• Gerard HB. Disagreement with others, their credibility, and experienced stress. Journal of Abnormal and Social Psychology. 1961;62:559–564. [PubMed] [Google Scholar]
• Gergen KJ, Jones EE. Mental illness, predictability, and affective consequences as stimulus factors in person perception. Journal of Abnormal and Social Psychology. 1963;67:95–105. [PubMed] [Google Scholar]
• Gilbert GM. Stereotype persistence and change among college students. Journal of Abnormal and Social Psychology. 1951;46:245–254. [PubMed] [Google Scholar]
• Glick P, Fiske ST. The Ambivalent Sexism Inventory: Differentiating hostile and benevolent sexism. Journal of Personality and Social Psychology. 1996;70:491–512. [Google Scholar]
• Goffman I. The presentation of self in everyday life. Garden City, NY: Doubleday; 1959. [Google Scholar]
• Greenberg BS, Miller GR. The effects of low credible sources on message acceptance. Speech Monographs. 1966;33:127–136. [Google Scholar]
• Grice HP. Logic and conversation. In: Cole P, Morgan JL, editors. Syntax and Semantics Vol. 3: Speech Acts. New York: Academic Press; 1975. pp. 41–58. [Google Scholar]
• Guilford JP. Racial preferences of a thousand American university students. Journal of Social Psychology. 1931;2:179–204. [Google Scholar]
• Harris A, Corner A, Hahn U. “Damned by faint praise”: A Bayesian account. In: Taatgen NA, van Rijn H, editors. Proceedings of the 31st Annual Conference of the Cognitive Science Society. Austin, TX: Cognitive Science Society; 2009. pp. 292–297. [Google Scholar]
• Hornsey MJ, Oppes T, Svensson A. “It’s OK if we say it, but you can’t”: Responses to intergroup and intragroup criticism. European Journal of Social Psychology. 2002;32:293–307. [Google Scholar]
• Jackman MR, Muha MJ. Education and intergroup attitudes: Moral enlightenment, superficial democratic commitment, or ideological refinement? American Sociological Review. 1984;49:751–769. [Google Scholar]
• Jones EE, Davis KE. From acts to dispositions: The attribution process in person perception. In: Berkowitz L, editor. Advances in experimental social psychology. Vol. 2. New York, NY: Academic Press; 1965. pp. 219–266. [Google Scholar]
• Jones EE, Hester SL, Farina A, Davis KE. Reactions to unfavorable personal evaluations as a function of the evaluator’s perceived adjustment. Journal of Abnormal and Social Psychology. 1959;59:63–370. [PubMed] [Google Scholar]
• Judd CM, James-Hawkins L, Yzerbyt V, Kashima Y. Fundamental dimensions of social judgment: Understanding the relations between judgments of competence and warmth. Journal of Personality and Social Psychology. 2005;89:899–913. [PubMed] [Google Scholar]
• Kanouse DE, Hanson LR., Jr . Negativity in evaluations. In: Jones EE, Kanouse DE, Kelley HH, Nisbett RE, Valins S, Weiner B, editors. Attribution: Perceiving the causes of behavior. Morristown, NJ: General Learning Press; 1972. pp. 47–62. [Google Scholar]
• Kaplan KJ. On the ambivalence-indifference problem in attitude theory and measurement: A suggested modification of the semantic differential technique. Psychological Bulletin. 1972;77:361–372. [Google Scholar]
• Karlins M, Coffman TL, Walters G. On the fading of social stereotypes: Studies in three generations of college students. Journal of Personality and Social Psychology. 1969;13:1–16. [PubMed] [Google Scholar]
• Katz D, Allport FH. Students’ attitudes. Syracuse, NY: Craftsman Press; 1931. [Google Scholar]
• Katz D, Braly K. Racial stereotypes of one hundred college students. Journal of Abnormal and Social Psychology. 1933;28:280–290. [Google Scholar]
• Katz D, Braly KW. Racial prejudice and racial stereotypes. Journal of Abnormal and Social Psychology. 1935;30:175–193. [Google Scholar]
• Kervyn N, Bergsieker HB, Fiske ST. The innuendo effect: Hearing the positive but inferring the negative. Journal of Experimental Social Psychology. 2012;48:77–85. [PMC free article] [PubMed] [Google Scholar]
• Kervyn N, Yzerbyt VY, Demoulin S, Judd CM. Competence and warmth in context: The compensatory nature of stereotypic views of national groups. European Journal of Social Psychology. 2008;38:1175–1183. [Google Scholar]
• Kervyn N, Yzerbyt VY, Judd CM. Compensation between warmth and competence: Antecedents and consequences of a negative relation between the two fundamental dimensions of social perception. European Review of Social Psychology. 2010;21:155–187. [Google Scholar]
• Lerner JS, Tetlock PE. Accounting for effects of accountability. Psychological Bulletin. 1999;129:255–275. [PubMed] [Google Scholar]
• Lewicka M, Czapinski J, Peeters G. Positive-negative asymmetry or “when the heart needs a reason” European Journal of Social Psychology. 1992;22:425–434. [Google Scholar]
• Lewis LD, Darley JM, Glucksberg S. Stereotype persistence and change among college students. Princeton University; Princeton, NJ: 1972. [Google Scholar]
• Lin MH, Kwan VSY, Cheung A, Fiske ST. Stereotype content model explains prejudice for an envied outgroup: Scale of Anti-Asian American Stereotypes. Personality and Social Psychology Bulletin. 2005;31:34–47. [PubMed] [Google Scholar]
• Lyons A, Kashima Y. How are stereotypes maintained through communication? The influence of stereotype sharedness. Journal of Personality and Social Psychology. 2003;85:989–1005. [PubMed] [Google Scholar]
• Madera JM, Hebl MR, Martin RC. Gender and letters of recommendation for academia: Agentic and communal differences. Journal of Applied Psychology. 2009;94:1591–1599. [PubMed] [Google Scholar]
• Madon S, Guyll M, Aboufadel K, Montiel E, Smith A, Palumbo P, Jussim L. Ethnic and national stereotypes: The Princeton Trilogy revisited and revised. Personality and Social Psychology Bulletin. 2001;27:996–1010. [Google Scholar]
• Mae L, Carlston DE. Hoist on your own petard: When prejudiced remarks are recognized and backfire on speakers. Journal of Experimental Social Psychology. 2005;41:240–255. [Google Scholar]
• Matlin MW, Stang DJ. The Pollyanna Principle: Selectivity in language, memory, and thought. Cambridge, MA: Schenkman; 1978. [Google Scholar]
• McConahay JB. Modern racism, ambivalence, and the Modern Racism Scale. In: Dovidio JF, Gaertner SL, editors. Prejudice, discrimination, and racism. New York: Academic Press; 1986. pp. 91–125. [Google Scholar]
• Meertens RW, Pettigrew TF. Is subtle prejudice really prejudice? Public Opinion Quarterly. 1997;61:54–71. [Google Scholar]
• Mercer J. Too marvelous for words: The magic of Johnny Mercer. Secaucus, NJ: Warner Brothers Publications; 1985. [Google Scholar]
• Merriam-Webster’s collegiate dictionary. 11. Springfield, MA: Merriam-Webster; 2005. [Google Scholar]
• Monin B, Miller DT. Moral credentials and the expression of prejudice. Journal of Personality and Social Psychology. 2001;81:33–43. [PubMed] [Google Scholar]
• Monteith MJ, Deneen NE, Tooman GD. The effect of social norm activation on the expression of opinions concerning gay men and blacks. Basic and Applied Social Psychology. 1996;18:267–288. [Google Scholar]
• Peeters G. Evaluative meanings of adjectives in vitro and in context: Some theoretical implications and practical consequences of positive negative asymmetry and behavioural-adaptive concepts of evaluation. Psychologia Belgica. 1992;32:211–231. [Google Scholar]
• Plant EA, Devine PG. Internal and external motivation to respond without prejudice. Journal of Personality and Social Psychology. 1998;75:811–832. [Google Scholar]
• Preacher KJ, Hayes AF. Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods. 2008;40:879–891. [PubMed] [Google Scholar]
• Price KO, Harburg E, Newcomb TM. Psychological balance in situations of negative interpersonal attitudes. Journal of Personality and Social Psychology. 1966;3:265–270. [PubMed] [Google Scholar]
• Rogers RW, Prentice-Dunn S. Deindividuation and anger-mediated interracial aggression: Unmasking regressive racism. Journal of Personality and Social Psychology. 1981;41:63–73. [Google Scholar]
• Rosenberg MJ. The conditions and consequences of evaluation apprehension. In: Rosenthal R, Rosnow RL, editors. Artifact in behavioral research. New York: Academic Press; 1969. pp. 279–349. [Google Scholar]
• Russell AMT, Fiske ST. It’s all relative: Competition and status drive interpersonal perception. European Journal of Social Psychology. 2008;38:1193–1201. [Google Scholar]
• Saguy T, Tausch N, Dovidio JF, Pratto F. The irony of harmony: Intergroup contact can produce false expectations for equality. Psychological Science. 2009;20:114–121. [PubMed] [Google Scholar]
• Schaller M, Conway LG., III Influence of impression-management goals on the emerging contents of group stereotypes: Support for a social-evolutionary process. Personality and Social Psychology Bulletin. 1999;25:819–833. [Google Scholar]
• Schaller M, Conway LG, III, Tanchuk TL. Selective pressures on the once and future contents of ethnic stereotypes: Effects of the communicability of traits. Journal of Personality and Social Psychology. 2002;82:861–877. [PubMed] [Google Scholar]
• Sears DO. The person-positivity bias. Journal of Personality and Social Psychology. 1983;44:233–250. [Google Scholar]
• Simon L, Greenberg J. Further progress in understanding the effects of derogatory ethnic labels: The role of preexisting attitudes toward the targeted group. Personality and Social Psychology Bulletin. 1996;22:1195–1204. [Google Scholar]
• Skowronski JJ, Carlston DE, Mae L, Crawford MT. Spontaneous trait transference: Communicators take on the qualities they describe in others. Journal of Personality and Social Psychology. 1998;74:837–848. [PubMed] [Google Scholar]
• Smith ER, Collins EC. Contextualizing person perception: Distributed social cognition. Psychological Review. 2009;116:343–364. [PubMed] [Google Scholar]
• Stangor C, Sechrist GB, Jost JT. Changing racial beliefs by providing consensus information. Personality and Social Psychology Bulletin. 2001;27:486–496. [Google Scholar]
• Sutton RM, Elder TJ, Douglas KM. Reactions to internal and external criticism of outgroups: Social convention in the intergroup sensitivity effect. Personality and Social Psychology Bulletin. 2006;32:563–575. [PubMed] [Google Scholar]
• Thorndike EL. A constant error in psychological ratings. Journal of Applied Psychology. 1920;4:25–29. [Google Scholar]
• Thurstone LL. An experimental study of nationality preferences. Journal of General Psychology. 1928;1:405–425. [Google Scholar]
• Wojciszke B. Morality and competence in person- and self-perception. In: Stroebe W, Hewstone M, editors. European Review of Social Psychology. Vol. 16. New York: Taylor & Francis; 2005. pp. 155–188. [Google Scholar]
• Wojciszke B, Bazinska R, Jaworski M. On the dominance of moral categories in impression formation. Personality and Social Psychology Bulletin. 1998;24:1251–1263. [Google Scholar]
• Wyer RS, Budesheim TL, Lambert AJ. Cognitive representation of conversations about persons. Journal of Personality and Social Psychology. 1990;58:218–238. [PubMed] [Google Scholar]
• Yzerbyt VY, Kervyn N, Judd CM. Compensation versus halo: The unique relations between the fundamental dimensions of social judgment. Personality and Social Psychology Bulletin. 2008;34:1110–1123. [PubMed] [Google Scholar]